No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research
Abstract
Artificial neural networks (ANNs) are biologically inspired computer programs designed to simulate the way in which the human brain processes information. ANNs gather their knowledge by detecting the patterns and relationships in data and learn (or are trained) through experience, not from programming. An ANN is formed from hundreds of single units, artificial neurons or processing elements (PE), connected with coefficients (weights), which constitute the neural structure and are organised in layers. The power of neural computations comes from connecting neurons in a network. Each PE has weighted inputs, transfer function and one output. The behavior of a neural network is determined by the transfer functions of its neurons, by the learning rule, and by the architecture itself. The weights are the adjustable parameters and, in that sense, a neural network is a parameterized system. The weighed sum of the inputs constitutes the activation of the neuron. The activation signal is passed through transfer function to produce a single output of the neuron. Transfer function introduces non-linearity to the network. During training, the inter-unit connections are optimized until the error in predictions is minimized and the network reaches the specified level of accuracy. Once the network is trained and tested it can be given new input information to predict the output. Many types of neural networks have been designed already and new ones are invented every week but all can be described by the transfer functions of their neurons, by the learning rule, and by the connection formula. ANN represents a promising modeling technique, especially for data sets having non-linear relationships which are frequently encountered in pharmaceutical processes. In terms of model specification, artificial neural networks require no knowledge of the data source but, since they often contain many weights that must be estimated, they require large training sets. In addition, ANNs can combine and incorporate both literature-based and experimental data to solve problems. The various applications of ANNs can be summarised into classification or pattern recognition, prediction and modeling. Supervised 'associating networks can be applied in pharmaceutical fields as an alternative to conventional response surface methodology. Unsupervised feature-extracting networks represent an alternative to principal component analysis. Non-adaptive unsupervised networks are able to reconstruct their patterns when presented with noisy samples and can be used for image recognition. The potential applications of ANN methodology in the pharmaceutical sciences range from interpretation of analytical data, drug and dosage form design through biopharmacy to clinical pharmacy.
... Limitation: Fluid flow/behavior not taken into consideration A novel approach using ANN and numerical wellbore simulation successfully predicted pressure drops in geothermal wells. ANN are computer programs (much like the human brain) that is modelled to learn from data patterns and relationships rather than explicit programming [1]. A range of ANN models were evaluated, all utilizing the Levenberg-Marquardt algorithm and either hyperbolic tangent sigmoid and linear transfer functions. ...
... I. replacing the ANN approach with a regression model [1]. This serves to achieve a similar outcome but with a different, but simple, yet accurate method of prediction of the value of dependent variables. ...
The use of fossil fuel for power generation releases large amounts of carbon dioxide into the atmosphere. The average world temperature is steadily rising, the greenhouse effect continues to intensify and the atmospheric carbon dioxide concentration has also surpassed 400 ppm. According to the World Health Organization, climate change is expected to cause approximately hundreds of thousands additional deaths per year between 2030 and 2050. Considering this, there is the need for a transition for the adoption of renewable energy sources. Geothermal energy, being one of the renewable energy sources, is a safe method of power generation but it runs the risk of triggering seismic activities if the pressure of the geothermal reservoir is not well-regulated. This project provides a C# model aimed at regulating the geothermal reservoir pressure using the production and injection rates. The C# model was produced using the Visual Studio 2022 software with the .NET framework and ML.NET packages from Microsoft. A sample data with three sets of parameters; the rate at which the geothermal fluid is extracted from the reservoir, the rate at which the fluid is injected back and their corresponding geothermal pressure value, was used for the implementation of the model. With the data fed into the model, rate of production as well as injection can increased or decreased based on the plant's requirement and the appropriate pressure for that change will be predicted by the C# model. The technique used for the prediction of geothermal reservoir pressure by the implemented model was regression, with the help of the Lbfgs Poisson Regression algorithm. The model was evaluated using statistical models like R2, mean squared error and mean absolute error. The input quantities of production and injection rates were set to 755 kg/s and 1500 kg/s respectively with the predicted pressure value being 4312.5786 psi. The R2 was 0.9839 which seemed to be close to a perfect fit (which is 1). Simultaneously, the mean absolute error was calculated to be 71.2759 psi and the mean squared error was 6758.4955 psi which represented a significant presence of error and this was drastically reduced or improved to 0.03275 psi and 0.00143 psi. using Microsoft Excel. The 2 model carries out great prediction and is reliable. The proposed model provides a simple yet accurate means of prediction using regression in C#. The predictive model can be modified to incorporate several other parameters like temperature and/or even permeability.
... (2) Artificial Neural Network An artificial neural network (ANN) is modeled based on biological neural networks [34]. Every neural network consists of three essential components: node features, network topology, and learning rules. ...
... In this study, to eliminate scale differences among different features, all input/output data (i.e., reflectance and SPAD) were scaled linearly to the range of [−1, 1]. Moreover, due to the impact of water vapor [34], some measured reflectance values at certain wavelengths may be considered outliers. Therefore, it is also important to deal with the outliers in the reflectance data. ...
Camellia oleifera is an oilseed crop that holds significant economic, ecological, and social value. In the realm of Camellia oleifera cultivation, utilizing hyperspectral analysis techniques to estimate chlorophyll content can enhance our understanding of its physiological parameters and response characteristics. However, hyperspectral datasets contain information from many wavelengths, resulting in high-dimensional data. Therefore, selecting effective wavelengths is crucial for processing hyperspectral data and modeling in retrieval studies. In this study, by using hyperspectral data and chlorophyll content from Camellia oleifera samples, three different dimensionality reduction methods (Taylor-CC, NCC, and PCC) are used in the first round of dimensionality reduction. Combined with these methods, various thresholds and dimensionality reduction methods (with/without further dimensionality reduction) are used in the second round of dimensionality reduction; different sets of core wavelengths with equal size are identified respectively. Using hyperspectral reflectance data at different sets of core wavelengths, multiple machine learning models (Lasso, ANN, and RF) are constructed to predict the chlorophyll content of Camellia oleifera. The purpose of this study is to compare the performance of various dimensionality reduction methods in conjunction with machine learning models for predicting the chlorophyll content of Camellia oleifera. Results show that (1) the Taylor-CC method can effectively select core wavelengths with high sensitivity to chlorophyll variation; (2) the two-stage hybrid dimensionality reduction methods demonstrate superiority in three models; (3) the Taylor-CC + NCC method combined with an ANN achieves the best predictive performance of chlorophyll content. The new two-stage dimensionality reduction method proposed in this study not only improves both the efficiency of hyperspectral data processing and the predictive accuracy of models, but can serve as a complement to the study of Camellia oleifera properties using the Taylor-CC method.
... We employed three machine learning models: random forest (RF), 15 support vector machine (SVM), 16 and artificial neural networks (ANN). 17 The RF model achieved a sensitivity of 0.95, a specificity of 0.77, and an AUC of 0.94. The SVM, using a radial basis function (RBF) kernel with C = 1 and gamma = 0.1, achieved a sensitivity of 0.94, a specificity of 0.75, and an AUC of 0.92. ...
... This approach accelerates materials discovery and design by identifying promising candidates for specific applications. A study by Agatonovic-Kustrin and Beresford (2000) discusses the principles of ANNs and their applications in predicting material properties. The authors demonstrate how ANNs can be trained to predict the behavior of materials under different conditions, providing a reliable and efficient method for materials characterization. ...
... During this experiment, four baseline dam monitoring prediction methods were used for evaluation and comparison to test the model generalization efect. Tese baseline methods include multiple linear regression (MR), and various ML-based methods, including backpropagation neural network (BPNN) [30], support vector regression (SVR) [31], and XGBoost [32]. Note that these benchmark methods use the same input variables as the developed method to validate its efectiveness. ...
Deformation is a critical indicator for the safety control of high-arch dams, yet traditional statistical regression methods often exhibit poor predictive performance when applied to long-sequence time series data. In this study, we develop a robust predictive model for deformation behavior in high-arch dams by integrating signal dimensionality reduction with deep learning (DL)-based residual correction techniques. First, the fast Fourier transform is employed to decompose air and water temperature sequences, enabling the extraction of temperature cycle characteristics at the dam boundary. A data-driven statistical monitoring model for dam deformation, based on actual temperature data, is then proposed. Subsequently, an improved Bayesian Ridge regression model is used to construct the dam deformation monitoring framework. The residuals that traditional statistical methods fail to capture are input into an enhanced Long Short-Term Memory (LSTM) network to effectively learn the temporal characteristics of the sequence. A high-arch dam with a history of long-term service is used as a case study. Experimental results indicate that the data dimensionality reduction method effectively extracts relevant information from observed temperature data, reducing the number of input variables. Comparative evaluation experiments show that the proposed hybrid predictive model outperforms existing state-of-the-art benchmark algorithms in terms of predictive efficiency and accuracy. Additionally, this approach combines the interpretability of statistical regression methods with the powerful nonlinear modeling capabilities of DL-based models, achieving a synergistic effect.
... The basic ANN model[9]. ...
Clay soil has undesirable engineering properties, which can compromise structural stability. This study aims to enhance the compaction properties of high-plasticity clay soil by adding glass powder using artificial intelligence (AI), specifically through the Back propagation Neural Network (BPNET), to accurately predict dry density. The model used influential factors, such as wet soil weight (Wnet), glass powder ratio (Wglass), and water content (ω %) as inputs, with dry density (γ) as the output. The model demonstrated high accuracy, achieving a Mean Squared Error (MSE) of 0.0000117 and a Mean Absolute Error (MAE) of 0.002849, reflecting its effectiveness in improving clay soil properties and supporting its stability.
... An artificial neural network (ANN) is a computational model inspired by the brain's structure and functions, designed to process information by emulating biological neuron operations [20]. ANNs comprise multiple layers of neurons; each receives inputs, processes them, and transmits the results to subsequent layers. ...
Arc droop presents significant challenges in power system management due to its inherent complexity and dynamic nature. To address these challenges in predicting arc sag for transmission lines, this paper proposes an innovative time–series prediction model, AROA-CNN-LSTM-Attention(AROA-CLA). The model aims to enhance arc sag prediction by integrating a convolutional neural network (CNN), a long short-term memory network (LSTM), and an attention mechanism, while also utilizing, for the first time, the adaptive rabbit optimization algorithm (AROA) for CLA parameter tuning. This combination improves both the prediction performance and the generalization capability of the model. By effectively leveraging historical data and exhibiting superior time–series processing capabilities, the AROA-CLA model demonstrates excellent prediction accuracy and stability across different time scales. Experimental results show that, compared to traditional and other modern optimization models, AROA-CLA achieves significant improvements in RMSE, MAE, MedAE, and R2 metrics, particularly in reducing errors, accelerating convergence, and enhancing robustness. These findings confirm the effectiveness and applicability of the AROA-CLA model in arc droop prediction, offering novel approaches for transmission line monitoring and intelligent power system management.
... After the procedure is nished, a test dataset with comparable features is utilized with the trained model. The ANN model known as the Multi-layer Perceptron (MLP) [44] is represented by Eq. 1: ...
Anthrax is a highly lethal disease caused by Bacillus anthracis. Lethal factor (LF) with protective antigen directly contributes to anthrax symptoms in humans. This research work identified a small molecule inhibitors of anthrax lethal factor. We developed a consolidated computational strategy that includes a deep learning-based SMOTE + artificial neural network (ANN) hybrid model, principal component analysis, t-SNE, activity cliff, constellation plot, scaffold, and fingerprinting to identify potential drug candidates against Anthrax. The best model showed 0.98 accuracy, 0.99 specificity, 0.99 sensitivity, 0.99 F1-score, 0.99 recall, 0.99 ROC, and 0.99 precision. The trained hybrid model screened out 134 FDA-approved drugs, 338 experimental drugs, 51 phytochemical compounds of the phytochemical database, and eight natural products from NCI divest IV as anthrax inhibitors. We found scaffold of ring system with substitution patterns such as 4-oxopyrrolo[3,2-c]quinolone enhanced the biological activity of Anthrax inhibitors. Fingerprints indicated greater than 80% and are linked to the ring system using the substitution pattern scaffold. These studies conclude that SMOTE + ANN model could be an efficient method for the virtual screening of large database and a new way to screen small molecules against Anthrax.
... 3. Theory and model OECD (2019) published a technical report artificial intelligence can help shape policies for improving lives. Agatonovic-Kustrin and Beresford (2000) applied artificial neural network (ANN) modelling in pharmaceutical research. Artificial intelligence encompasses intelligent systems to learn with reasoning and to perceive with action. ...
Purpose
This research explores support vector machine (SVM) with Gaussian radial basis function kernel (RBF) as the model and Analysis of Variance (ANOVA) for forecasting the invalidation re-examination decisions of China invention patents, it is beneficial to support patent monetization for corporate intellectual capital.
Design/methodology/approach
There were 8,666 China invention patents with their existing invalidation re-examination decisions during 2000∼2021 chosen to conduct classification model training and prediction for the accuracy of invalidation re-examination decisions through SVM with RBF. Statistical significance was performed by ANOVA to identify indicators for these invention patents selected in this research. These selected 8,666 China invention patents were divided into two groups based on their invalidation re-examination decisions during 2000∼2021 in Table 1, which Group 1 included 5,974 invention patents with all valid or partially valid claims, and Group 0 included 2,692 invention patents with all invalid claims. Thereafter, each group was further divided into sub-groups based on 13 major regions where the applicants filed invalidation re-examination. The training sets for Group 1, Group 0 and the sub-groups were selected based on the patent issued in January, February, April, May, July, August, October and November; while the prediction sets were selected from the invention patents issued in March, June, September and December.
Findings
The training and prediction accuracies were compared to the existing invalidation re-examination decisions. Accuracies of training sets were ranged from 100% in region 7 (Beijing) and region 9 (Shanghai) to 95.95% in region 1 (US), and the average accuracy of invalidation re-examination decisions was 98.95%. While the accuracies of prediction sets for Group 1 were ranged from 100.00% in region 7 (Beijing) to 90.78% in region 13 (Overseas-others), and the average accuracy of classification was 95.96%, this research’s outcomes confirmed the purpose of applying SVM with RBF to predict the patentability sustainability.
Originality/value
This research developed an empirical method through SVM with RBF to predict patentability sustainability which is crucial for corporate intellectual capital on patents. In particular, the investments on patents are huge, including the patent cultivation and maintenance, developments into products or services, patent litigations and dispute managements. Therefore, this research is beneficial not only for corporation, but also for research organisations to perform cost-effective and profitable patent strategies on intellectual capital.
... Especially under the condition of AT, the ML-based method may be efficient for OAM demultiplexing/recognition. Various neural network approaches (NNA), including Artificial Neural Networks (ANN) [19], Generative Adversarial Networks (GAN) [20], Deep Neural Networks (DNN) [21], and Convolutional Neural Networks (CNN) [22], have been developed for OAM recognition, with CNN being the most prevalent. In 2019, Wang et al investigated the accuracy of measuring OAM using CNN under different transmission distances in AT and various superposition states of vortex beams with different OAM mode intervals [23]. ...
Vortex light carrying orbital angular momentum (OAM) is a beam with a helical phase structure. The OAM light has great potential in the field of communication, due to the fact that it can greatly improve the efficiency and capacity of information transmission. One of the popular information propagation methods is coding and decoding by different single OAM light or combined multiple OAM lights. However, Laguerre–Gaussian (LG) beams (LGB) carrying OAM, which is a classical vortex light, are prone to distortion of intensity under atmospheric turbulence (AT) disturbances. Due to the influence of AT, the effective recognition of the OAM mode in a free space becomes an important challenge for information propagation. For mitigating the influence of AT, a scheme combining AT extraction and OAM modes recognition is proposed, which can efficiently identify both AT intensity and OAM modes. A 99 % identification accuracy of AT can be reached by the proposed scheme. Besides, the obtained results also show that the recognition rate of OAM modes is greatly improved after the introduction of AT extraction module, especially under strong turbulence conditions. Compared to direct-mode-identification method without extracting AT, the recognition accuracy can be improved by 8 % and 3 % when the AT intensity is 1×10−13 and 5×10−14m−2/3 , respectively. Consequently, the proposed scheme can be used to identify the OAM modes with a high accuracy, which is beneficial to OAM coding and decoding in an OAM-based communication system.
... Comprising interconnected nodes arranged in layers (input, hidden, and output), this model employs weighted connections and activation functions to facilitate information processing. The network iteratively adjusts weights throughout training to minimize the disparity between predicted and actual outputs [46]. Furthermore, this study employed a fivefold cross-validation technique to minimize errors and reduce the risk of model overfitting [44,45]. ...
Background
Stroke is a significant global health concern, ranking as the second leading cause of death and placing a substantial financial burden on healthcare systems, particularly in low- and middle-income countries. Timely evaluation of stroke severity is crucial for predicting clinical outcomes, with standard assessment tools being the Rapid Arterial Occlusion Evaluation (RACE) and the National Institutes of Health Stroke Scale (NIHSS). This study aims to utilize Machine Learning (ML) algorithms to predict stroke severity using these two distinct scales.
Methods
We conducted this study using two datasets collected from hospitals in Urmia, Iran, corresponding to stroke severity assessments based on RACE and NIHSS. Seven ML algorithms were applied, including K-Nearest Neighbor (KNN), Decision Tree (DT), Random Forest (RF), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Artificial Neural Network (ANN). Hyperparameter tuning was performed using grid search to optimize model performance, and SHapley Additive Explanations (SHAP) were used to interpret the contribution of individual features.
Results
Among the models, the RF achieved the highest performance, with accuracies of 92.68% for the RACE dataset and 91.19% for the NIHSS dataset. The Area Under the Curve (AUC) was 92.02% and 97.86% for the RACE and NIHSS datasets, respectively. The SHAP analysis identified triglyceride levels, length of hospital stay, and age as critical predictors of stroke severity.
Conclusions
This study is the first to apply ML models to the RACE and NIHSS scales for predicting stroke severity. The use of SHAP enhances the interpretability of the models, increasing clinicians’ trust in these ML algorithms. The best-performing ML model can be a valuable tool for assisting medical professionals in predicting stroke severity in clinical settings.
... • MLP We use the traditional neural network multi-layer perceptron to solve this classification task [54]. ...
Causal inference and model interpretability are gaining increasing attention, particularly in the biomedical domain. Despite recent advance, decorrelating features in nonlinear environments with human-interpretable representations remains underexplored. In this study, we introduce a novel method called causal rule generation with target trial emulation framework (CRTRE), which applies randomize trial design principles to estimate the causal effect of association rules. We then incorporate such association rules for the downstream applications such as prediction of disease onsets. Extensive experiments on six healthcare datasets, including synthetic data, real-world disease collections, and MIMIC-III/IV, demonstrate the model's superior performance. Specifically, our method achieved a error of 0.907, outperforming DWR (1.024) and SVM (1.141). On real-world datasets, our model achieved accuracies of 0.789, 0.920, and 0.300 for Esophageal Cancer, Heart Disease, and Cauda Equina Syndrome prediction task, respectively, consistently surpassing baseline models. On the ICD code prediction tasks, it achieved AUC Macro scores of 92.8 on MIMIC-III and 96.7 on MIMIC-IV, outperforming the state-of-the-art models KEPT and MSMN. Expert evaluations further validate the model's effectiveness, causality, and interpretability.
... The emergence and advancement of deep learning have sparked a growing interest in the utilization of artificial neural networks (ANN) for solving partial differential equations. 2 Deep learning methods, such as physics-informed neural networks (PINNs), have emerged as versatile approaches for solving partial differential equations. 3-5 PINN utilizes neural networks to approximate the solution of the desired PDE and incorporate residual terms into the loss function. ...
Physics-informed neural networks (PINNs) have become a promising research direction in the field of solving partial differential equations (PDEs). Dealing with singular perturbation problems continues to be a difficult challenge in the field of PINN. The solution of singular perturbation problems often exhibits sharp boundary layers and steep gradients, and traditional PINN cannot achieve approximation of boundary layers. In this manuscript, we propose the General-Kindred physics-informed neural network (GKPINN) for solving singular perturbation differential equations (SPDEs). This approach utilizes asymptotic analysis to acquire prior knowledge of the boundary layer from the equation and establishes a novel network to assist PINN in approximating the boundary layer. It is compared with traditional PINN by solving examples of one-dimensional, two-dimensional, and time-varying SPDE equations. The research findings underscore the exceptional performance of our novel approach, GKPINN, which delivers a remarkable enhancement in reducing the L2 error by two to four orders of magnitude compared to the established PINN methodology. This significant improvement is accompanied by a substantial acceleration in convergence rates, without compromising the high precision that is critical for our applications. Furthermore, GKPINN still performs well in extreme cases with perturbation parameters of 1×10−38, demonstrating its excellent generalization ability.
... Recent advancements in the field of neural networks have led to an era where the computational efficiency of SNNs is being harnessed in more innovative and practical ways [1,44,46]. SNNs, with their ability to mimic the intricacies of biological neural networks, represent a significant leap from traditional artificial neural networks. They offer a more nuanced approach to information processing, mimicking the dynamic, temporal characteristics of biological neuron activity. ...
Tiny Machine Learning (TinyML) has become a growing field in on-device processing for Internet of Things (IoT) applications, capitalizing on AI algorithms that are optimized for their low complexity and energy efficiency. These algorithms are designed to minimize power and memory footprints, making them ideal for the constraints of IoT devices. Within this domain, Spiking Neural Networks (SNNs) stand out as a cutting-edge solution for TinyML, owning to their event-driven processing paradigm which offers an efficient method of handling dataflow. This paper presents a novel SNN architecture based on the 1st Order Leaky Integrate-and-Fire (LIF) neuron model to efficiently deploy vision-based ML algorithms on TinyML systems. A hardware-friendly LIF design is also proposed, and implemented on a Xilinx Artix-7 FPGA. To evaluate the proposed model, a collision avoidance dataset is considered as a case study. The proposed SNN model is compared to the state-of-the-art works and Binarized Convolutional Neural Network (BCNN) as a baseline. The results show the proposed approach is 86% more energy efficient than the baseline.
... The theoretical foundation of the single neural layer ANN machine learning model is described in earlier research [28]. In hydrology, ANN is widely used and has proven competent in various prediction tasks. ...
To develop agricultural risk management strategies, the early identification of water deficits during the growing cycle is critical. This research proposes a deep learning hybrid approach for multi-step soil moisture forecasting in the Bundaberg region in Queensland, Australia, with predictions made for 1-day, 14-day, and 30-day, intervals. The model integrates Geospatial Interactive Online Visualization and Analysis Infrastructure (Giovanni) satellite data with ground observations. Due to the periodicity, transience, and trends in soil moisture of the top layer, time series datasets were complex. Hence, the Maximum Overlap Discrete Wavelet Transform (moDWT) method was adopted for data decomposition to identify the best correlated wavelet and scaling coefficients of the predictor variables with the target top layer moisture. The proposed 3-phase hybrid moDWT-Lasso-LSTM model used the Least Absolute Shrinkage and Selection Operator (Lasso) method for feature selection. Optimal hyperparameters were identified using the Hyperopt algorithm with deep learning LSTM method. This proposed model's performances were compared with benchmarked machine learning (ML) models. In total, nine models were developed, including three standalone models (e.g., LSTM), three integrated feature selection models (e.g., Lasso-LSTM), and three hybrid models incorporating wavelet decomposition and feature selection (e.g., moDWT-Lasso-LSTM). Compared to alternative models, the hybrid deep moDWT-Lasso-LSTM produced the superior predictive model across statistical performance metrics. For example, at 1-day forecast, The moDWT-Lasso-LSTM model exhibits the highest accuracy with the highest R 2 ≈ 0.92469 and the lowest RMSE ≈ 0.97808, MAE ≈ 0.76623, and SMAPE ≈ 4.39700%, outperforming other models. The moDWT-Lasso-DNN model follows closely, while the Lasso-ANN and Lasso-DNN models show lower accuracy with higher RMSE and MAE values. The ANN and DNN models have the lowest performance, with higher error metrics and lower R2 values compared to the deep learning models incorporating moDWT and Lasso techniques. This research emphasizes the utility of the advanced complementary ML model, such as the developed moDWT-Lasso-LSTM 3-phase hybrid model, as a robust data-driven tool for early forecasting of soil moisture.
... Artificial Neural Networks (ANNs) have emerged as a pivotal technology in the field of artificial intelligence (AI), driving advancements across numerous domains including computer vision, natural language processing, and autonomous systems. Inspired by the intricate networks of neurons in the human brain, ANNs are designed to simulate the way biological systems process information, offering a powerful framework for addressing complex computational tasks that traditional algorithms struggle to handle [6][7][8]. The resurgence of interest in ANNs over the past decade can be attributed to several key developments. ...
Implementing Artificial Neural Networks (ANNs) on Field-Programmable Gate Arrays (FPGAs) provides a promising solution for achieving high-performance, low-latency, and energy-efficient computations in complex tasks. This paper investigates the methodology for mapping ANNs onto FPGAs, focusing on critical aspects such as architecture selection, hardware design, and optimization techniques. By harnessing the parallel processing capabilities and reconfigurability of FPGAs, neural network computations are significantly accelerated, making them ideal for real-time applications like image processing and embedded systems. The implementation process addresses key considerations, including fixed-point arithmetic, memory management, and dataflow optimization, while employing advanced techniques such as pipelining, quantization, and pruning. The research compares the accuracy and performance speedup of ANNs on CPUs versus FPGAs, revealing that FPGA-based simulations are 4680 times faster than CPU-based simulations using MATLAB, without compromising prediction accuracy. INTRODUCTION Artificial intelligent is now a driving force behind various technologies used in websites, cameras, and smartphones. It is often implemented to identify objects in images and extract them for further processing. The general approach of machine learning involves taking a real dataset and applying algorithms such as deep learning [1], neural networks [2], the Perceptron algorithm [3], K-nearest neighbor [4], decision trees [5], among others. Among these, ANN has emerged as one of the most dominant techniques. ANNs excel in extracting and analyzing data, allowing them to effectively establish relationships between inputs and outputs.
... The theoretical foundation of the single neural layer ANN machine learning model is described in earlier research [28]. In hydrology, ANN is widely used and has proven competent in various prediction tasks. ...
To develop agricultural risk management strategies, the early identification of water deficits during the growing cycle is critical. This research proposes a deep learning hybrid approach for multi-step soil moisture forecasting in the Bundaberg region in Queensland, Australia, with predictions made for 1-day, 14-day, and 30-day, intervals. The model integrates Geospatial Interactive Online Visualization and Analysis Infrastructure (Giovanni) satellite data with ground observations. Due to the periodicity, transience, and trends in soil moisture of the top layer, time series datasets were complex. Hence, the Maximum Overlap Discrete Wavelet Transform (moDWT) method was adopted for data decomposition to identify the best correlated wavelet and scaling coefficients of the predictor variables with the target top layer moisture. The proposed 3-phase hybrid moDWT-Lasso-LSTM model used the Least Absolute Shrinkage and Selection Operator (Lasso) method for feature selection. Optimal hyperparameters were identified using the Hyperopt algorithm with deep learning LSTM method. This proposed model's performances were compared with benchmarked machine learning (ML) models. In total, nine models were developed, including three standalone models (e.g., LSTM), three integrated feature selection models (e.g., Lasso-LSTM), and three hybrid models incorporating wavelet decomposition and feature selection (e.g., moDWT-Lasso-LSTM). Compared to alternative models, the hybrid deep moDWT-Lasso-LSTM produced the superior predictive model across statistical performance metrics. For example, at 1-day forecast, The moDWT-Lasso-LSTM model exhibits the highest accuracy with the highest R 2 ≈ 0.92469 and the lowest RMSE ≈ 0.97808, MAE ≈ 0.76623, and SMAPE ≈ 4.39700%, outperforming other models. The moDWT-Lasso-DNN model follows closely, while the Lasso-ANN and Lasso-DNN models show lower accuracy with higher RMSE and MAE values. The ANN and DNN models have the lowest performance, with higher error metrics and lower R2 values compared to the deep learning models incorporating moDWT and Lasso techniques. This research emphasizes the utility of the advanced complementary ML model, such as the developed moDWT-Lasso-LSTM 3-phase hybrid model, as a robust data-driven tool for early forecasting of soil moisture.
... This combination forms the input to an activation function, introducing non-linearity which allows the neural network to effectively model the intricate relationships between inputs and outputs. The network adjusts these weights and biases to minimize prediction error [27]. ...
... ANN are computational models that contain hundreds of single units, artificial neurons, connected with coefficients (weights) in which the human's brain working principles are emulated to conduct learning and then the prediction. ANNs can be trained to model a complex problem with many parameters where the training process is performed using proper exemplars [18]. ANN consists of interconnected neurons that are processing elements having similar characteristics, such as inputs, synaptic strength, activation outputs, and bias [19]. ...
Steel beam-column connections with shape memory alloy (SMA) bolts provide self-centering behavior and eliminate permanent deformations in earthquake-resilient steel moment frames. This paper presents the development and frame modeling application of a freely available Graphical User Interface (GUI) for predicting the cyclic and self-centering response of extended endplate steel connections with superelastic SMA bolts. A database of moment-rotation response is created from the results of seventy-two 3D finite-element simulations and seven experimentally tested specimens. The study trains different machine learning algorithms, including Artificial Neural Networks (ANN), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosted Trees (ExGBT), Light Gradient Boosted Trees (LGBT), TensorFlow Deep Learning (TFDL), and Keras Deep Residual Neural Network (KDP). The accuracy of the algorithms is compared in terms of two performance metrics, root mean square error (RMSE) and coefficient of determination (R 2). The trained ANNs, which show the highest accuracy with an R 2 ranging from 0.92 to 0.99, are chosen to develop a Graphical User Interface (GUI). To demonstrate the application of the developed predictive tool, a phenomenological model of moment frames with SMA connections is developed and verified in OpenSees based on experimental test results. A two-stage validation study is performed to assess the accuracy of the proposed phenomenological model. The validation study uses the predictive tool to develop the phenomenological model for SMA-based beam-to-column connection. It is shown that the results of the predictive tool, the phenomenological model, and the 3D finite-element models in ANSYS are in line with each other with high accuracy. By using the developed tool, the prediction of the cyclic and self-centering response of a typical SMA connection can be performed rapidly-taking three minutes in OpenSees compared to seven hours in ANSYS.
... The use of Artificial Neural Networks in the modeling processes of dynamic systems has long drained the currents of scientific research [5][6][7][8]. This trend takes its rise in the elegance of neural models, more particularly their ability to model, using basic electronic components, more complex phenomena and behaviors. ...
We consider a dynamical system consisting of a van der Pol oscillator linearly coupled to an inertial neuron with two wells potential. Analytical studies are conducted focusing on the energy computation, the dissipation and symmetry, as well as the determination and characterization of the equilibrium points. We define the parameter ranges related to different types of oscillations in the coupled system in order to have an overall idea of the nature of the attractors (hidden or self-excited) that may exist. We apply numerical analysis techniques (2-parameter diagrams, bifurcation analysis, phase portraits, basins of attractions, etc) in accordance with the previous operating range in order to shed light on the plethora of competing dynamics of the model and possible forms of strange attractors as well. Another salient point of this work is the coexistence between five self-excited attractors (limit cycle and chaos) with a hidden attractor (limit cycle). We also examine the impact of symmetry breaking on the system response. An appropriate analog simulator of the coupled system is designed and simulated in PSpice in order to check the results reported during the theoretical analyses. We believe that the results of the present work complement and enrich previously published ones concerning the dynamics of a system composed of a van der pol oscillator coupled to a (non-oscillating) double-well oscillator.
... Artificial neural networks (ANNs) are computational frameworks inspired by the biological neurons in brains (Agatonovic-Kustrin & Beresford, 2000;Burke, 1991). An ANN consists of artificial neurons (units), each with a weight (coefficient) connected with neurons in the next layer. ...
Machine learning (ML) techniques have shown promising applications in a broad range of topics in engineering, composite materials behavior analysis, and manufacturing. This paper reviews successful ML implementations for defect and damage identification and progression in composites. The focus is on predicting composites' responses under specific loads and environments and optimizing setting and imperfection sensitivity. Discussions and recommendations toward promising ML implementation practices for fruitful interpretable results in the composites’ analysis are provided.
... Finally, the output layer encompasses all pieces of data processed by the network and can return the response to the environment. This layer can be made from singular or agglomerated nodes 33 . ...
The cast in-situ pile is a widely used type of deep foundations which its execution in civil projects is increasing daily. The use of ordinary concrete in this type of piles causes technical and executive problems, a decrease in the compressive strength (CS) of concrete, and an increase in the permeability under the ground level. But use of the self-compacting concrete in the cast in-situ piles while increasing the CS of concrete ensures proper compaction, increase in the execution speed, and easy placing of concrete. In this article, utilizing the data obtained from the laboratory results and also the application of soft computing techniques, predicting the degree of CS of self-compacting concrete (SCC) in concrete piles was investigated. To estimate the CS of SCC, a total number of 7 inputs were implemented. Then, using gene expression programming (GEP) a model was presented for estimating the CS of SCC in the cast in-situ piles. The results of the neural network showed a precision of 99.98% which exhibits the high accuracy of the model. The use of this model could greatly help persons, companies, and research centers in the preparation and construction of self-compacting concrete with the desired CS.
... Every ANN architecture contains an input layer, one or more hidden layers and an output layer (Abdulrahman et al., 2020). Transfer functions used at the output layer determine the type of output generated by the ANN (Agatonovic- Kustrin & Beresford, 2000). An illustration of a basic ANN architecture is shown in Fig. 4. ...
Additive manufacturing (AM), also commonly known as 3D Printing, has emerged as a revolutionary technique for manufacturing components having complex geometries. Fused Deposition Modelling (FDM) is one of the most commonly used methods in additive manufacturing due to multiple factors namely, availability of compatible materials, design ease, reduced wastage etc. However, FDM-based additive manufacturing suffers from some critical issues such as increased build time, need for the support structure, lower mechanical strength, sub-optimal dimensional accuracy, surface roughness etc. In recent times, researchers have attempted to improve these output characteristics and overcome some of these limitations with the help of computational intelligence-based methods. Specifically, supervised machine learning-based techniques have proven to be quite useful in model design, quality evaluation as well as in-situ monitoring. This paper explores and outlines opportunities for such applications and reports specific learnings published over the past few years. Analytical insights into the selection of ML models, model training, and current challenges have been presented to guide further research and experimentation in this interdisciplinary area.
Digital marketing significantly influences travellers’ decision-making processes. This research aims to analyze the specific digital marketing strategies affecting Gen-Z travellers’ decisions. Machine learning techniques were employed to uncover key insights. Data were collected through a survey using a convenience random sampling of 346 university students in Thailand. Correlation analysis and machine learning approaches, such as logistic regression and neural network analysis, were used to explore the relationships between various digital marketing strategies and their impact on Gen-Z travellers’ destination choices. The findings indicate that Online Travel Agency (OTA) websites exert the most significant influence on Gen-Z travellers’ decision-making processes, with an odds ratio of 1.6879. This is followed by the use of social media marketing strategies and businesses having their own websites. To validate these results, an Artificial Neural Network (ANN) technique was also employed, confirming the importance of rankings derived from the logistic regression analysis. This research provides valuable insights into which digital marketing strategies most influence Gen-Z travellers’ decision-making processes. These insights can help tourism businesses make informed investments in digital marketing strategies, prioritizing those that align with the preferences of Gen-Z travellers.
AI Innovations in Drug Delivery and Pharmaceutical Sciences: Advancing Therapy through Technology offers a comprehensive exploration of how artificial intelligence (AI) is revolutionizing the pharmaceutical and healthcare sectors. This book addresses the AI’s role in drug discovery, development, and delivery, highlighting applications in personalized medicine, nanotechnology, and clinical trials. It also covers AI’s impact on community and hospital pharmacy, herbal medicine, and drug product design. Each chapter examines the use of AI in optimizing drug processes, from designing innovative therapies to improving regulatory compliance and future trends in pharmaceutical technology. This insightful resource is invaluable for researchers, pharmaceutical professionals, and healthcare innovators aiming to advance therapeutic outcomes through AI. Key Features: - Comprehensive coverage of AI applications in drug discovery, delivery, and design. - Insights into AI-driven personalized medicine and nanotechnology. - Regulatory perspectives on AI in drug delivery and medical devices. - Future trends and innovations in AI for pharmaceutical technology.
As life expectancy and the proportion of the population over the age of 65 increases, osteoporosis and its clinical consequence, fragility fractures, has become a growing issue in both medical and economic terms. With the objective of facilitating early diagnosis and forecasting the outcomes of osteoporosis and its repercussions, the global scientific community has initiated efforts to develop artificial intelligence (AI)-based solutions. Specifically, artificial intelligence (AI) models that use either supervised or unsupervised learning have particularly compelling applications in screening for osteoporosis and fragility fractures, evaluating fracture risk, assessing response to treatment, and data analysis in applied research. However, the capabilities of this technology as an augment to medical practice should not be overestimated, as it has not been able to outperform conventional approaches for fracture prediction. Better model design is required to ensure appropriate application and validation of AI in the treatment of osteoporosis.
General circulation models (GCM) represent a contemporary and advanced tool designed to simulate the response of climate system to alterations in greenhouse gas levels. Increasing spatial resolutions of the outputs of GCMs on a regional scale requires downscaling process. In this study, four machine learning (ML) models, namely, decision tree regression (DTR), support vector regression, artificial neural network, and K-nearest neighbors (KNN) were employed to downscale the outputs of four Coupled Model Intercomparison Project 6 climate models. The daily observations of temperature data at Nazmakan station in Kohgiluyeh and Boyer-Ahmad Province, Iran, were trained and tested for the periods 1995–2009 and 2009–2015, respectively. According to results of four metrics, DTR and KNN obtained the most valid evaluation in the training and test data, respectively. In addition, downscaling methods were used to predict future daily temperature for 2015–2045 under two Shared Socioeconomic Pathway (SSP2-4.5 and SSP5-8.5) scenarios. Mann–Kendall test results revealed an increasing trend of temperature in most of the coming years. The projected trend expects cooler summers and warmer winters in future. Furthermore, the predicted data under two climate scenarios were assessed in comparison with observed data for 2015–2022. Finally, this study demonstrates the strengths and weaknesses of nonlinear ML techniques in statistical downscaling.
India is highly vulnerable to atmospheric pollution, especially in city areas, due to the exponential growth in urbanization and industrialization. The increasing level of pollutants in the atmosphere worsened the ambient air quality. The Air Quality Index (AQI) has been increasing at an alarming rate in major cities of India. This dire situation has helped us shed more light on the modelling of ground-level ozone formation over major Indian cities like Delhi and Bangalore. Ground-level ozone is a secondary pollutant formed from various precursors like NOx and VOCs in the presence of sunlight. Pollutant forecasting will be helpful in providing valuable information for selecting the optimal air pollution control strategies. Artificial Neural Networks (ANN) performed better than linear models and Support Vector Machines (SVM) regression models for the modelling of tropospheric ozone formation in both cities. The R² values for linear models were 0.757 and 0.8116, and for SVM regression models, R² values were 0.79 and 0.86, respectively, for Delhi and Bangalore. Conversely, the ANN models for Delhi and Bangalore yielded R² values of 0.91 and 0.95, respectively. This clearly depicts the ANN model’s efficacy compared to linear and SVM models for modelling tropospheric ozone formation. Sensitivity analysis has revealed the importance of meteorological parameters for ozone formation in Delhi and Bangalore. Also, the importance of meteorological parameters varied for both cities. Solar radiation has the highest impact in Delhi, while in Bangalore, it is relative humidity.
To overcome the challenges of limited experimental data and improve the accuracy of empirical formulas, we propose a low-cycle fatigue (LCF) life prediction model for nickel-based superalloys using a data augmentation method. This method utilizes a variational autoencoder (VAE) to generate low-cycle fatigue data and form an augmented dataset. The Pearson correlation coefficient (PCC) is employed to verify the similarity of feature distributions between the original and augmented datasets. Six machine learning models, namely random forest (RF), artificial neural network (ANN), support vector machine (SVM), gradient-boosted decision tree (GBDT), eXtreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost), are utilized to predict the LCF life of nickel-based superalloys. Results indicate that the proposed data augmentation method based on VAE can effectively expand the dataset, and the mean absolute error (MAE), root mean square error (RMSE), and R-squared (R2) values achieved using the CatBoost model, with respective values of 0.0242, 0.0391, and 0.9538, are superior to those of the other models. The proposed method reduces the cost and time associated with LCF experiments and accurately establishes the relationship between fatigue characteristics and LCF life of nickel-based superalloys.
Purpose
In the existing era, international trade is boosted by maritime freight movement. The academicians and Government are concerned about environmental contamination caused by maritime goods that transit global growth and development. Digital technologies like blockchain help the maritime freight business to stay competitive in the digital age. This study aims to illuminate blockchain technology (BCT) adoption aspects to alleviate early industry adoption restrictions.
Design/methodology/approach
This study adopts a two-stage approach comprising of structural equation modeling (SEM) with artificial neural networks (ANN) to analyze critical factors influencing the adoption of BCT in the sustainable maritime freight industry.
Findings
The SEM findings from this study illustrate that social, organizational, technological and infrastructual and institutional factors affect BCT execution. Furthermore, the ANN technique uses the SEM data to determine that sustainability enabled digital freight training (S3), initial investment cost (O5) and trust over digital technology (G1) are the most essential blockchain deployment factors.
Originality/value
The hybrid approach aims to help decision-makers and policymakers examine their organizational blockchain adoption goals to construct sustainable, efficient and effective maritime freight transportation.
The use of fossil fuel for power generation releases large amounts of carbon dioxide into the atmosphere. The average world temperature is steadily rising, the greenhouse effect continues to intensify and the atmospheric carbon dioxide concentration has also surpassed 400 ppm. According to the World Health Organization, climate change is expected to cause approximately hundreds of thousands additional deaths per year between 2030 and 2050. Considering this, there is the need for a transition for the adoption of renewable energy sources. Geothermal energy, being one of the renewable energy sources, is a safe method of power generation but it runs the risk of triggering seismic activities if the pressure of the geothermal reservoir is not well-regulated. This project provides a C# model aimed at regulating the geothermal reservoir pressure using the production and injection rates. The C# model was produced using the Visual Studio 2022 software with the .NET framework and ML.NET packages from Microsoft. A sample data with three sets of parameters; the rate at which the geothermal fluid is extracted from the reservoir, the rate at which the fluid is injected back and their corresponding geothermal pressure value, was used for the implementation of the model. With the data fed into the model, rate of production as well as injection can increased or decreased based on the plant's requirement and the appropriate pressure for that change will be predicted by the C# model. The technique used for the prediction of geothermal reservoir pressure by the implemented model was regression, with the help of the LbfgsPoissonRegression algorithm. The model was evaluated using statistical models like R2, mean squared error and mean absolute error. The input quantities of production and injection rates were set to 755 kg/s and 1500 kg/s respectively with the predicted pressure value being 4312.5786 psi. The R2 was 0.9839 which seemed to be close to a perfect fit (which is 1). Simultaneously, the mean absolute error was calculated to be 71.2759 psi and the mean squared error was 6758.4955 psi which represented a significant presence of error and this was drastically reduced or improved to 0.03275 psi and 0.00143 psi. using Microsoft Excel. The 2 model carries out great prediction and is reliable. The proposed model provides a simple yet accurate means of prediction using regression in C#. The predictive model can be modified to incorporate several other parameters like temperature and/or even permeability.
Algorithms predicting RNA secondary structures based on different folding criteria – minimum free energies (mfe), kinetic
folding (kin), maximum matching (mm) – and different parameter sets are studied systematically. Two base pairing alphabets
were used: the binary GC and the natural four-letter AUGC alphabet. Computed structures and free energies depend strongly on both the algorithm and the parameter set. Statistical
properties, such as mean number of base pairs, mean numbers of stacks, mean loop sizes, etc., are much less sensitive to the
choice of parameter set and even of algorithm. Some features of RNA secondary structures, such as structure correlation functions,
shape space covering and neutral networks, seem to depend only on the base pairing logic (GC or AUGC alphabet).
The certainty-factor (CF) model is a commonly used method for managing uncertainty in rule-based systems. We review the history and mechanics of the CF model, and delineate precisely its theoretical and practical limitations. In addition, we examine the belief network, a representation that is similar to the CF model but that is grounded firmly in probability theory. We show that the belief-network representation overcomes many of the limitations of the CF model, and provides a promising approach to the practical construction of expert systems.
Random graph theory is used to model relationships between sequences and secondary structure of RNA molecules. Sequences folding into identical structures form neutral networks which percolate sequence space if the fraction of neutral nearest neighbors exceeds a threshold value. The networks of any two different structures almost touch each other, and sequences folding into almost all "common" structures can be found in a small ball of an arbitrary location in sequence space. The results from random graph theory are compared with data obtained by folding large samples of RNA sequences. Differences are explained in terms of RNA molecular structures.
An artificial neural network system is used for pattern recognition in protein side-chain-side-chain contact maps. A back-propagation
network was trained on a set of patterns which are popular in side-chain contact maps of protein structures. Several neural
network architectures and different training parameters were tested to decide on the best combination for the neural network.
The resulting network can distinguish between original (from protein structures) and randomized patterns with an accuracy
of 84.5% and a Matthews' coefficient of 0.72 for the testing set. Applications of this system for protein structure evaluation
and refinement are also proposed. Examples include structures obtained after the application of molecular dynamics to crystal
structures, structures obtained from X-ray crystallography at various stages of refinement, structures obtained from a de novo folding algorithm and deliberately misfolded structures.
The quantitative similarity among tRNA gene sequences was acquired by analysis with an artificial neural network. The evolutionary relationship derived from our results was consistent with those from other methods. A new sequence was recognized to be a tRNA-like gene by a neural network on the analysis of similarity. All of our results showed the efficiency of the artificial neural network method in the sequence analysis for biological molecules.
The sequence patterns of 11 types of frequently occurring connecting peptides, which lead to a classification of supersecondary motifs, were studied. A database of protein supersecondary motifs was set up. An artificial neural network method, i.e. the back propagation neural network, was applied to the predictions of the supersecondary motifs from protein sequences. The prediction correctness ratios are higher than 70%, and many of them vary from 75 to 82%. These results are useful for the further study of the relationship between the structure and function of proteins. It may also provide some important information about protein design and the prediction of protein tertiary structure.
In the last decade, two tools, one drawn from information theory and the other from artificial neural networks, have proven particularly useful in many different areas of sequence analysis. The work presented herein indicates that these two approaches can be joined in a general fashion to produce a very powerful search engine that is capable of locating members of a given nucleic acid sequence family in either local or global sequence searches. This program can, in turn, be queried for its definition of the motif under investigation, ranking each base in context for its contribution to membership in the motif family. In principle, the method used can be applied to any binding motif, including both DNA and RNA sequence families, given sufficient family size.
The aim of this study was to develop a multi-objective simultaneous optimization technique in which an artificial neural network (ANN) was incorporated. As model formulations, 18 kinds of Trapidil tablet were prepared. The amounts of microcrystalline cellulose, hydroxypropyl methylcellulose and compression pressure were selected as causal factors. In order to characterize the release profiles of Trapidil, the release order and the rate constant were estimated. A set of release parameters and causal factors was used as tutorial data for ANN and fed into a computer. Non-linear relationships between causal factors and the release parameters were represented well with the response surface of ANN. The simultaneous optimization of the sustained-release tablet containing Trapidil was performed by minimizing the generalized distance between the predicted values of each response and the optimized one that was obtained individually. The optimal formulations gave satisfactory release profiles, since the observed results coincided well with the predicted results. These findings demonstrate that a multi-objective optimization technique incorporating ANN is quite useful in the optimization of pharmaceutical formulations.
An improved chemometric approach is proposed for assessing chromatographic peak purity by means of artificial neural networks. A non-linear transformation function with a back-propagation algorithm was used to describe and predict the chromatographic data. The mann-Whitney U-test was used for the concluding the purity of the chromatographic peak. Simulation data and practical analytical data for both pure and mixture samples were analysed with satisfactory results. A prior knowledge of the impurity and the related compound is unecessary when a slight difference between their chromatogram and spectrum exists. The performance on simulated data sets by this approach was compared with the results from principal component analysis.
Recent work on neural networks in chemistry is reviewed and essential background to this fast-spreading method is given. Emphasis is placed on the back-propagation algorithm, because of the extensive use of this form of learning. Hopfield networks, adaptive bidirectional associative memory, and Kohonen learning are briefly described and discussed. Applications in spectroscopy (mass, infrared, ultraviolet, NMR), potentiometry, structure/activity relationships, protein structure, process control and chemical reactivity are summarized.
The effect of the topology and the training end point of artificial neural networks (ANN) in the modelling of a fluidized bed granulation process is presented. The neural network topologies were designed on the basis of an earlier study (Murtoniemi et al., Int. J. Pharm., 108 (1994) 155–163). In the first part of this study, the networks contained only one hidden layer in which the number of neurons was either 10, 15, 20 or 25. The training end points with all four networks ranged from 0.15 to 0.07, with a step length of 0.01. In the second part, the training end point was fixed to be 0.12, while the number of neurons in the hidden layer varied from 10 to 25. The main purpose of this study was to find a suitable ANN in regard to the generalization ability and to compare the results to those calculated on the basis of multilinear stepwise regression analysis. The results showed that the number of hidden layer neurons did not affect the generalization ability of the networks and a proper generalization ability was achieved with rather simple networks. The training end point, however, had a significant effect on the generalization ability and it also affects the number of iteration epochs needed. In complicated systems this probably will affect remarkably the time required for the training.
The influence of different ratios of individual components (silicone surfactant, hydrophilic and lipophilic phase) on the viscoelastic behaviour of semisolid lipophilic emulsion systems was studied. The creams were prepared according to a preliminary experimental design (mixture design). The content of all three phases was varied: surfactant (1–5%), purified water (from 40 to 90%) and white petrolatum (5–59%). Oscillatory rheometry was used as the most appropriate experimental method for the evaluation of the emulsions. The rheological properties were influenced by the ratio of the components. For highly concentrated systems the predominant elastic response in the whole frequency range was measured. The cross-over point is characteristic for the concentrated systems. For the low concentrated systems, viscous behaviour is predominant. Rheometry has also been employed to follow and evaluate physical stability as one of the critical factors of emulsion systems. The dynamic rheological parameter, tan δ, has been chosen as the basis for developing mathematical models (linear, polynomial, neural) to forecast stability related to the content of the separate components in emulsion systems. After testing the linear models their non agreement to statistical criteria for a good model Freg, Flof, CC, DC, and RMS was found. The two-level neural network has been proven to be a statistically acceptable model, as were the polynomials of the second order. The two-level neural network model was also evaluated and the results have shown a great degree of reliability. The prediction of tan δ using a neural network model was found to be of great interest for the contemporary pharmaceutical formulation design because a lot of additional testing can be omitted.
The usefulness of artificial neural networks for response surface modeling in HPLC optimization is compared with multiple regression methods. The results show that neural networks offer promising possibilities in HPLC method development. The predicted capacity factors of analytes were better to those obtained with multiple regression method.
The application of ANN in pharmaceutical development has been assessed using theoretical as well as typical pharmaceutical technology examples. The aim was to quantitatively describe the achieved data fitting and predicting abilities of the models developed with a view to using ANN in the development of solid dosage forms. The comparison between the ANN and a traditional statistical (i.e., response surface methodology, RSM) modeling technique was carried out using the squared correlation coefficient R2. Using a highly nonlinear arbitrary function the ANN models showed better fitting (R2 = 0.931 vs. R2 = 0.424) as well as predicting (R2 = 0.810 vs. R2 = 0.547) abilities. Experimental data from a tablet compression study were fitted using two types of ANN models (i.e., multilayer perceptrons and a hybrid network composed of a self-organising feature map joined to a multilayer perception). The achieved data fitting was comparable for the three methods (MLP R2 = 0.911, SOFM-MLP R2 = 0.850, and RSM R2 = 0.897). ANN methodology represents a promising modeling technique when applied to pharmaceutical technology data sets.
Neural computing technology is capable of solving problems involving complex pattern recognition. This technology is applied here to pharmaceutical product development. The most commonly used computational algorithm, the delta back-propagation network, was utilized to recognize the complex relationship between the formulation variables and the in vitro drug release parameters for a hydrophilic matrix capsule system. This new computational technique was also compared with the response surface methodology (RSM). Artificial neural network (ANN) analysis was able to predict the response values for a series of validation experiments more precisely than RSM. ANN may offer an alternative to RSM because it allows for the development of a system that can incorporate literature and experimental data to solve common problems in the pharmaceutical industry.
This study demonstrates the application of neural networks to predict the pharmacokinetic properties of beta-adrenoreceptor antagonists in humans. A congeneric series of 10 beta-blockers, whose critical pharmacokinetic parameters are well established, was selected for the study. An appropriate neural network system was constructed and tested for its ability to predict the pharmacokinetic parameters from the octanol/water partition coefficient (shake flask method), the pKa, or the fraction bound to plasma proteins. Neural networks successfully trained and the predicted pharmacokinetic values agreed well with the experimental values (average difference = 8%). The neural network-predicted values showed better agreement with the experimental values than those predicted by multiple regression techniques (average difference = 47%). Because the neural networks had a large number of connections, two tests were conducted to determine if the networks were memorizing rather than generalizing. The "leave-one-out" method verified the generalization of the networks by demonstrating that any of the compounds could be deleted from the training set and its value correctly predicted by the new network (average error = 19%). The second test involved the prediction of pharmacokinetic properties of compounds never seen by the network, and reasonable results were obtained for three out of four compounds tested. The results indicate neural networks can be a powerful tool in exploration of quantitative structure-pharmacokinetic relationships.
Genetic algorithms provide a novel tool for the investigation of combinatorial optimization problems. A genetic algorithm takes an initial set of possible starting solutions, and iteratively improves them by means of crossover and mutation operators that are related to those involved in Darwinian evolution. This approach is illustrated by reference to applications in molecular modelling, the docking of flexible ligands into protein active sites and de novo ligand design.
The usefulness of artificial neural networks for response surface modeling in HPLC optimization is compared with (non-)linear regression methods. The number of hidden nodes is optimized by a lateral inhibition method. Overfitting is controlled by cross-validation using the leave one out method (LOOM). Data sets of linear and non-linear response surfaces (capacity factors) were taken from literature. The results show that neural networks offer promising possibilities in HPLC method development. The predictive results were better or comparable to those obtained with linear and non-linear regression models.
A new hybrid method (GNN) combining a genetic algorithm and an artificial neural network has been developed for quantitative structure-activity relationship (QSAR) studies. A suitable set of molecular descriptors are selected by a genetic algorithm. This set serves as input to a neural network, in which model-free mapping of multivariate data is performed. Multiple predictors are generated that are superior to results obtained from previous studies of the Selwood data set, which is used to test the method. The neural network technique provides a graphical description of the functional form of the descriptors that play an important role in determining drug activity. This can serve as an aid in future design of drug analogues. The effectiveness of GNN is tested by comparing its results with a benchmark obtained by exhaustive enumeration. Different fitness strategies that tune the evolution of genetic models are examined, and QSARs with higher predictiveness are found. From these results, a composite model is constructed by averaging predictions from several high-ranking models. The predictions of the resulting QSAR should be more reliable than those derived from a single predictor because it makes greater use of information and also permits error estimation. An analysis of the sets of descriptors selected by GNN shows that it is essential to have one each for the steric, electrostatic, and hydrophobic attributes of a drug candidate to obtain a satisfactory QSAR for this data set. This type of result is expected to be of general utility in designing and understanding QSAR.
A novel model-independent approach to analyze pharmacokinetic (PK)-pharmacodynamic (PD) data using artificial neural networks (ANNs) is presented. ANNs are versatile computational tools that possess the attributes of adaptive learning and self-organization. The emulative ability of neural networks is evaluated with simulated PK-PD data, and the power of ANNs to extrapolate the acquired knowledge is investigated. ANNs of one architecture are shown to be flexible enough to accurately predict PD profiles for a wide variety of PK-PD relationships (e.g., effect compartment linked to the central or peripheral compartment and indirect response models). Also, an example is given of the ability of ANNs to accurately predict PD profiles without requiring any information regarding the active metabolite. Because structural details are not required, ANNs exhibit a clear advantage over conventional model-dependent methods. ANNs are proved to be robust toward error in the data and perturbations in the initial estimates. Moreover, ANNs were shown to handle sparse data well. Neural networks are emerging as promising tools in the field of drug discovery and development.
Preliminary investigations have been conducted to assess the potential for using artificial neural networks to simulate aerosol behaviour, with a view to employing this type of methodology in the evaluation and design of pulmonary drug-delivery systems.
Details are presented of the general purpose software developed for these tasks; it implements a feed-forward back-propagation algorithm with weight decay and connection pruning, the user having complete run-time control of the network architecture and mode of training. A series of exploratory investigations is then reported in which different network structures and training strategies are assessed in terms of their ability to simulate known patterns of fluid flow in simple model systems. The first of these involves simulations of cellular automata-generated data for fluid flow through a partially obstructed two-dimensional pipe. The artificial neural networks are shown to be highly successful in simulating the behaviour of this simple linear system, but with important provisos relating to the information content of the training data and the criteria used to judge when the network is properly trained. A second set of investigations is then reported in which similar networks are used to simulate patterns of fluid flow through aerosol generation devices, using training data furnished through rigorous computational fluid dynamics modelling. These more complex three-dimensional systems are modelled with equal success.
It is concluded that carefully tailored, well trained networks could provide valuable tools not just for predicting but also for analysing the spatial dynamics of pharmaceutical aerosols.
A novel method for deconvolving overlapped peaks in chromatograms is proposed. The basic idea of this method consists of finding a set of parameters which characterize the shape of the overlapped peaks and using a multi-layered perceptron network for quantitatively correlating the parameters with the percentage area of an individual peak. The proposed method performs very well with high accuracy and less computing time compared to other conventional methods.
Preliminary investigations have been conducted to assess the potential for using (back-propagation, feed-forward) artificial neural networks to predict the phase behavior of quaternary microemulsion-forming systems, with a view to employing this type of methodology in the evaluation of novel cosurfactants for the formulation of pharmaceutically acceptable drug-delivery systems. The data employed in training the neural networks related to microemulsion systems containing lecithin, isopropyl myristate, and water, together with different types of cosurfactants, including short- and medium-chain alcohols, amines, acids, and ethylene glycol monoalkyl ethers. Previously unpublished phase diagrams are presented for four systems involving the cosurfactants 2-methyl-2-butanol, 2-methyl-1-propanol, 2-methyl-1-butanol, and isopropanol, which, along with eight other published sets of data, are used to test the predictive ability of the trained networks. The pseudo-ternary phase diagrams for these systems are predicted using only four computed physicochemical properties for the cosurfactants involved. The artificial neural networks are shown to be highly successful in predicting phase behavior for these systems, achieving mean success rates of 96.7 and 91.6% for training and test data, respectively. The conclusion is reached that artificial neural networks can provide useful tools for the development of microemulsion-based drug-delivery systems.
The origins and operation of artificial neural networks are briefly described and their early application to data modelling in drug design is reviewed. Four problems in the use of neural networks in data modelling are discussed, namely overfitting, chance effects, overtraining and interpretation, and examples are given of the means by which the first three of these may be avoided. The use of neural networks as a variable selection tool is shown and the advantage of networks as a nonlinear data modelling device is discussed. The display of multivariate data in two dimensions employing a neural network is illustrated using experimental and theoretical data for a set of charge transfer complexes.
The ability of neural network models to predict aqueous solubility within series of structurally related drugs was evaluated. Three sets of compounds representing different drug classes (28 steroids, 31 barbituric acid derivatives, and 24 heterocyclic reverse transcriptase inhibitors) were studied. Topological descriptors (connectivity indices, kappa indices, and electrotopological state indices) were used to link the structures of compounds with their aqueous solubility. Separate models were built for each class of drugs using back-propagation neural networks with one hidden layer and five topological indices as input parameters. The effect of network size and training time on the prediction ability of the network was studied by the leave-one-out (LOO) procedure. In all three compound groups a neural network structure of 5-3-1 was optimal. To avoid chance effects, artificial neural network (ANN) ensembles (i.e.; averaging neural network predictions over several independent networks) were used. The cross-validated squared correlation coefficient (Q2) for 10 averaged predictions was 0.796 in the case of the steroid set. The corresponding standard error of prediction (SDEP) was 0.288 log units. For the barbiturates, Q2 and SDEP were 0.856 and 0.383, respectively, and for the RT inhibitors, these parameters were 0.721 and 0.401, respectively. The results indicate that neural networks can produce useful models of the aqueous solubility of a congeneric set of compounds, even with simple structural parameters.
Back-propagation artificial neural networks (ANNs) were trained with parameters derived from different molecular structure representation methods, including topological indices, molecular connectivity, and novel physicochemical descriptors to model the structure--activity relationship of a large series of capsaicin analogues. The ANN QSAR model produced a high level of correlation between the experimental and predicted data. After optimization, using cross-validation and selective pruning techniques, the ANNs predicted the EC50 values of 101 capsaicin analogues, correctly classifying 34 of 41 inactive compounds and 58 of 60 active compounds. These results demonstrate the capability of ANNs for predicting the biological activity of drugs, when trained on an optimal set of input parameters derived from a combination of different molecular structure representations.
Artificial neural networks provide a unique computing architecture whose potential has attracted interest from researchers across different disciplines. As a technique for computational analysis, neural network technology is very well suited for the analysis of molecular sequence data. It has been applied successfully to a variety of problems, ranging from gene identification, to protein structure prediction and sequence classification. This article provides an overview of major neural network paradigms, discusses design issues, and reviews current applications in DNA/RNA and protein sequence analysis.
The utility of genetic neural network (GNN) to obtain quantitative structure-activity relationships (QSAR) from molecular similarity matrices is described. In this application, the corticosteroid-binding globulin (CBG) binding affinity of the well-known steroid data set is examined. Excellent predictivity can be obtained through the use of either electrostatic or shape properties alone. Statistical validation using a standard randomization test indicates that the results are not due to chance correlations. Application of GNN on the combined electrostatic and shape matrix produces a six-descriptor model with a cross-validated r2 value of 0.94. The model is superior to those obtained from partial least-squares and genetic regressions, and it also compares favorably with the results for the same data set from other established 3D QSAR methods. The theoretical basis for the use of molecular similarity in QSAR is discussed.
The use of artificial neural networks (ANNs) for response surface modelling in HPLC method development for amiloride and methychlothiazide separation is reported. The independent input variables were pH and methanol percentage in mobile phase. The outputs were capacity factors. The results were compared with a statistical method (multiple nonlinear regression analysis). Networks were able to predict the experimental responses more accurately than the regression analysis.
A method for predicting the aqueous solubility of drug compounds was developed based on topological indices and artificial neural network (ANN) modeling. The aqueous solubility values for 211 drugs and related compounds representing acidic, neutral, and basic drugs of different structural classes were collected from the literature. The data set was divided into a training set (n = 160) and a randomly chosen test set (n = 51). Structural parameters used as inputs in a 23-5-1 artificial neural network included 14 atom-type electrotopological indices and nine other topological indices. For the test set, a predictive r2 = 0.86 and s = 0.53 (log units) were achieved.
In this work we carry out a study of pattern recognition to detect the microbiological activity in a group of heterogeneous compounds. The structural descriptors utilized are the topological connectivity indexes. The methods followed are stepwise linear discriminant analysis (linear analysis) and artificial neural network (nonlinear analysis). Although both methods are appropriate to differentiate between active and inactive compounds, the artificial neural network is, in this case, more adequate, since it shows in a test set a prediction success of 98%, versus 92% obtained with linear discriminant analysis.
The absorption of a drug compound through the human intestinal cell lining is an important property for potential drug candidates. Measuring this property, however, can be costly and time-consuming. The use of quantitative structure-property relationships (QSPRs) to estimate percent human intestinal absorption (%HIA) is an attractive alternative to experimental measurements. A data set of 86 drug and drug-like compounds with measured values of %HIA taken from the literature was used to develop and test a QSPR mode. The compounds were encoded with calculated molecular structure descriptors. A nonlinear computational neural network model was developed by using the genetic algorithm with a neural network fitness evaluator. The calculated %HIA (cHIA) model performs wells, with root-mean-square (rms) errors of 9.4%HIA units for the training set, 19.7%HIA units for the cross-validation (CV) set, and 16.0%HIA units for the external prediction set.
Based on in vitro solubility and in vivo permeability drugs can be divided into four groups: class 1 (high permeability, high solubility, HP:HS), class 2 (high permeability, low solubility, HP:LS), class 3 (low permeability, high solubility, LP:HS), and class 4 (low permeability, low solubility, LP:LS) (Amidon et al., 1995; Amidon, G.L., Lennernas, H., Shah, V.P., Olson, J.R., 1995. Pharm. Res. 12, 413-420). The high permeability boundary has been suggested to be 70% (Walter et al., 1996; Walter, E., Janich, S., Roessler, B.J., Hilfinger, J.H., Amidon, G., 1996. J. Pharm. Sci. 85, 1070-1076) or 90% (Hussain et al., 1997; Hussain, A.S., Kaus, L.C., Lesko, L.J., Williams, R.L., 1997. Eur. J. Pharm. Sci. 5 (Suppl. 2), S43-S44), and more recently was compromised by the FDA to 80% human intestinal absorption (Hussain, A.S., 1998. Information presented at the BCS and in vitro-in vivo correlations workshop. Frankfurt/M., Germany, March 1998). The biopharmaceutics classification system (BCS) is now being considered by the FDA to develop new regulatory guidance for bioequivalence studies (Hussain, 1998; Lesko, 1997; Lesko, L.J., 1997. Eur. J. Pharm. Sci. 5 (Suppl. 2), S42). Both properties, solubility and permeability, are being considered as fundamental to define the rate and extent of absorption of the active ingredient of a drug product. However, since both these properties are dependent ones, it may be questioned whether these are indeed sufficiently 'fundamental' or should be further unravelled.
One of the difficulties in the quantitative approach to designing pharmaceutical formulations is the difficulty in understanding the relationship between causal factors and individual pharmaceutical responses. Another difficulty is desirable formulation for one property is not always desirable for the other characteristics. This is called a multi-objective simultaneous optimization problem. A response surface method (RSM) has proven to be a useful approach for selecting pharmaceutical formulations. However, prediction of pharmaceutical responses based on the second-order polynomial equation commonly used in RSM, is often limited to low levels, resulting in poor estimations of optimal formulations. The aim of this review is to describe the basic concept of the multi-objective simultaneous optimization technique in which an artificial neural network (ANN) is incorporated. ANNs are being increasingly used in pharmaceutical research to predict the non-linear relationship between causal factors and response variables. The usefulness and reliability of this ANN approach is demonstrated by the optimization for ketoprofen hydrogel ointment as a typical numerical example, in comparison with the results obtained with a classical RSM approach.
A simple X-ray powder diffractometric method was developed for the qualitative and quantitative assay of the two crystalline modifications of ranitidine-HCl. The main purpose of the present work was to investigate if artificial neural networks (ANNs) could be applied in quantitative X-ray diffractometric analyses. The ANN approach was compared with a conventional mixture design method. The results obtained by the ANN had a smaller standard deviation and relative error and a better precision at lower concentrations. ANNs provide a simple alternative to conventional statistical modelling methods to identify the non-linear relationship without complex equations.
A new, simple, sensitive and rapid method was developed to analyse the polymorphic purity of crystalline ranitidine-HCI as a bulk drug and from a tablet formulation.
Diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy was combined with Artificial Neural Networks (ANNs) as a data modelling tool. A standard feed-forward network, with backpropagation rule and with single hidden layer architecture was chosen. Reduction and transformation of the spectral data enhanced the ANN performance and reduced the complexity of the ANNs model. Spectral intensities from 1738 wavenumbers were reduced into 173 averaged spectral values. These 173 values were used as inputs for the ANN. Following a sensitivity analysis the number of inputs was reduced to 30, or 35, these being the input windows which had most effect on the output of the ANN.
For the bulk drug assay, the ANN model had 30 inputs selected from a sensitivity analysis, one hidden layer, and two output neurons, one for the percentage of each ranitidine hydrochloride crystal form. The model could simultaneously distinguish between crystal forms and quantify them enabling the physical purity of the bulk drug to be checked. For the tablet assay, the ANN model had 173 averaged spectral values as the inputs, one hidden layer and five output neurons, two for the percentage of the two ranitidine hydrochloride crystal forms and three more outputs for tablet excipients and additives. The ANN was able to solve the problem of overlapping peaks and it successfully identified and quantified all components in tablet formulation with reasonable accuracy.
Some of the advantages over conventional analytical methods include simplicity, speed and good selectivity. The results from DRIFT spectral quantification study show the benefits of the neural network approach in analysing spectral data.
The aim of the present work was to develop a method for predicting the phase behaviour of four component systems consisting of oil, water and two surfactants from a limited number of screening experiments. Investigations were conducted to asses the potential of artificial neural networks (ANNs) with back-propagation training algorithm to predict the phase behaviour of four component systems. Three inputs only (percentages of oil and water and HLB of the surfactant blend) and four outputs (oil in water emulsion, water in oil emulsion, microemulsion, and liquid crystals containing regions) were used. Samples used for training represented about 15% of the sampling space within the tetrahedron body. The network was trained by performing optimization of the number and size of the weights for neuron interconnections. The lowest error was obtained with 15 hidden neurons and after 4,500 training cycles. The trained ANN was tested on validation data and had an accuracy of 85.2-92.9%. In most cases the errors in the prediction were confined to points lying along the boundaries of regions and for the extrapolated predictions outside the ANNs 'experience'. This approach is shown to be highly successful and the ANNs have proven to be a useful tool for the prediction of the phase behaviour of quaternary systems with less experimental effort.